A Stacked Ensemble Spyware Detection Model Using Hyper-Parameter Tuned Tree Based Classifiers

Spyware is a type of malware that is designed to infiltrate a device or steal personal information. Over the last decade, the number of people facing such dangers has risen from 12.4 million to 812.67 million. Since the spyware target platform has been enlarged, additional strategies using non-detectable approaches have been noticed. Traditional detection approaches are ineffective in this instance since they can only detect known assaults. Advanced behavior-based detection can aid this problem as well as it can contribute to the detection of zero-day attacks. As a result, this study proposes an ensemble stacking learning-based method for detecting Spyware, which contains four conventional machine learning tree based techniques. To avoid biases and overfitting, we apply grid search hyperparameter tuning approach. Also, we perform an efficient feature selection technique that shows similar accuracy with less data dimension. At first in our proposed model, recursive feature elimination is used by utilizing a decision tree estimator to generate an optimized number for reducing data dimensions. Secondly, three different feature ranker selects K best features using the optimized number from the previous step. Finally, a union process is done avoiding redundancy to generate the resampled dataset. The experimental result shows around 88% accuracy with only a 0.35% error rate. The test accuracy is also around 96%, hence the model is not overfitted.